Upscale AI Introduces New Networking Platform for AI Data Centers
Artificial intelligence infrastructure company Upscale AI has introduced a new networking platform designed to connect AI processors from different manufacturers within the same data center. The platform, called Token Fabric, aims to simplify the deployment of large AI computing systems and give operators greater flexibility when selecting hardware.
Announced on October 8, 2026, Token Fabric combines networking hardware, software and management tools into a unified architecture. The system is designed to connect processors within individual computing clusters and across larger data-center environments.
The launch addresses a growing challenge for AI infrastructure operators: building powerful computing systems without having to rely entirely on one processor manufacturer or manually integrate separate networking technologies.
What Is Token Fabric?
Token Fabric is an integrated networking architecture developed specifically for AI infrastructure. It combines the connections needed inside a computing cluster with the networks that link multiple clusters across a data center.
Traditional data-center deployments can require separate systems and software to connect different components. As AI computing environments become more complex, coordinating those systems can increase deployment time and operational costs.
Upscale AI says Token Fabric is designed to bring these functions together under a common software and management framework.
The company's goal is to help businesses deploy AI infrastructure that can accommodate different types of accelerators while simplifying network configuration, monitoring and operations.
Designed to Connect Different AI Chips
One of the platform's central features is support for heterogeneous computing environments. This means infrastructure operators can build systems that use processors from different suppliers instead of depending exclusively on a single hardware ecosystem.
AI data centers increasingly use specialized processors for model training, inference and other workloads. These processors may have different networking requirements, performance characteristics and software environments.
Token Fabric is designed to help connect these systems through an open, standards-based architecture.
For cloud providers and companies operating large AI clusters, this approach could provide additional flexibility when selecting computing hardware for different applications.
Nvidia Technology Plays a Key Role
Upscale AI is backed by Nvidia and other investors. Its new platform incorporates systems powered by Nvidia Spectrum-X networking technology for connections across the data center.
The platform also includes Upscale's own SkyFabriX switching silicon for connecting accelerators within computing clusters.
These components are brought together through the company's SkyOS networking software and SkyCMD orchestration tools.
The approach combines Nvidia-based networking systems with Upscale's own technology while aiming to support infrastructure that includes different types of AI processors.
Why AI Data Centers Need Better Networking
Modern AI systems require more than powerful processors. They also depend on high-speed communication between processors, memory systems, storage and other computing resources.
When processors cannot exchange information efficiently, expensive computing hardware may spend time waiting for data instead of completing useful work.
Networking therefore plays an important role in the overall performance of AI infrastructure.
As companies deploy larger models and more demanding AI applications, they need systems that can transfer data quickly while coordinating thousands of connected components.
Token Fabric is designed to address this challenge by combining networking hardware and software into a more unified architecture.
SkyFabriX Targets High-Bandwidth Connections
Upscale AI's SkyFabriX technology is designed for high-bandwidth connections between accelerators inside AI computing clusters.
The company says its current switching silicon delivers 115.2 terabits per second of aggregate bandwidth, with a longer-term roadmap targeting multi-petabit-per-second capabilities.
High-bandwidth switching is important because AI workloads often require processors to exchange large amounts of information during training and inference.
The company is also developing systems intended to support open networking standards, giving infrastructure operators more options when designing future computing clusters.
These performance figures are company specifications and development targets; actual system performance will depend on deployment conditions, hardware configuration and workload requirements.
New Software to Manage AI Networks
Token Fabric includes two important software components: SkyOS and SkyCMD.
SkyOS provides the networking software foundation for the platform, while SkyCMD is designed to provide orchestration, monitoring and management across connected infrastructure.
By bringing these capabilities together, Upscale aims to make it easier for operators to manage large AI networks through a unified interface.
This could be particularly useful for cloud providers and enterprises that operate multiple AI clusters and need to monitor network performance, identify bottlenecks and coordinate infrastructure changes.
Upscale AI Has Raised $500 Million
Upscale AI has attracted significant investment as demand grows for specialized AI networking infrastructure.
The company was valued at approximately $2 billion in June 2026, when it announced a $190 million funding extension. Its investors include Nvidia, Salesforce, Temasek and Premji Invest, the investment firm associated with Indian technology entrepreneur Azim Premji.
The company has raised approximately $500 million in total funding, according to reporting and company information available around the launch.
The investment reflects growing interest in technologies that help data centers connect and operate large numbers of AI processors efficiently.
Cloud Providers Could Benefit From Greater Hardware Choice
Cloud computing companies and specialized AI infrastructure providers are building facilities to serve customers developing and running advanced AI models.
Many operators want to use different processors depending on the workload, cost and availability of hardware. However, integrating multiple computing platforms can introduce networking and software complexity.
A common networking architecture could help reduce that complexity by providing a more consistent way to connect and manage different systems.
For enterprises, the potential benefits include greater hardware flexibility and more options when planning future AI infrastructure investments.
These benefits will ultimately depend on compatibility, real-world performance, deployment costs and the availability of supported hardware.
Availability Is Planned for 2027
Upscale AI says general availability for the Token Fabric platform is planned for early 2027. Early-access and joint-validation programs are already underway with customers and ecosystem partners.
The company plans to offer different adoption options, allowing customers to select individual hardware components, integrated systems, software or a more complete networking solution.
This flexibility is intended to serve a range of customers, from large cloud providers that build their own infrastructure to businesses that prefer integrated systems.
The planned rollout means the platform's wider commercial adoption and real-world performance will become clearer as deployments progress.
Competition in AI Networking Is Increasing
AI networking is becoming an important part of the global semiconductor and data-center industry.
Companies are competing to build faster switches, more efficient interconnects and software that keeps AI processors working together effectively. Networking performance can affect how efficiently expensive computing hardware is used.
Upscale AI is entering this market with an approach that combines its own switching technology with Nvidia networking products and a unified software layer.
The broader industry is also exploring open standards that could make it easier to combine processors and networking equipment from different suppliers.
What Token Fabric Could Mean for the Future of AI
The next generation of AI infrastructure will require coordination between processors, memory, storage, networking and software. As computing clusters grow, managing these components as separate systems can become increasingly difficult.
Token Fabric aims to address this problem by bringing multiple networking functions into one architecture while supporting different types of AI accelerators.
If the platform delivers on its stated goals, it could give data-center operators more flexibility when selecting hardware and expanding AI computing capacity.
However, its long-term impact will depend on successful deployments, interoperability with other technologies and whether customers adopt the system at scale.
The AI Infrastructure Race Moves Beyond Chips
The introduction of Token Fabric highlights an important development in the artificial intelligence industry: competition is no longer limited to designing faster processors.
Companies also need efficient networks capable of connecting large numbers of chips and allowing them to operate as coordinated computing systems.
Upscale AI is attempting to make that infrastructure more flexible through a combination of switching silicon, networking systems and software.
With commercial availability planned for early 2027, the platform will be one to watch as cloud providers and enterprises look for new ways to build and manage increasingly complex AI data centers.
Journalist: Vijay Singh